Nvidia Unveils Rubin Chip Architecture Promising Major AI Advancements
Nvidia will soon produce its new Rubin architecture, boasting enhanced AI capabilities and speed for major cloud providers.
Huang explained that Vera Rubin is here to help with the huge need for AI computation. He noted, “Today, I can tell you that Vera Rubin is in full production.” The Rubin architecture is the latest product of Nvidia’s ongoing efforts to improve its hardware. It will take the place of the older Blackwell architecture, which replaced Hopper and Lovelace architectures before it.
Rubin chips are expected to be used by almost every major cloud provider, with key partnerships including Anthropic, OpenAI, and Amazon Web Services. They will also be used in HPE’s Blue Lion supercomputer and the new Doudna supercomputer at Lawrence Berkeley National Lab.
Named after the famous astronomer Vera Rubin, this architecture includes six chips that work together. The main chip is the Rubin GPU, but there are also improvements in storage and connections with the new Bluefield and NVLink systems. A new Vera CPU is included, which is designed for smart reasoning.
Dion Harris, Nvidia’s senior director of AI infrastructure solutions, talked about the new storage benefits. He said that modern AI systems need more memory every day. “We’ve introduced a new tier of storage that connects externally to the compute device,” Harris explained. This upgrade helps to manage the storage needed for AI tasks more efficiently.
As expected, the Rubin architecture is much faster and more energy-efficient. Nvidia’s tests show it runs three and a half times faster than Blackwell for training models and five times faster for inference tasks, reaching up to 50 petaflops. The new platform also allows for eight times more inference compute per watt of energy.
These upgrades come at a time when there is fierce competition to create better AI tools. Both AI labs and cloud companies are racing to acquire Nvidia chips and the necessary systems to support them. During an earnings call in October 2025, Huang estimated that between $3 trillion and $4 trillion will be spent on AI infrastructure over the next five years.